Scaling limits for point random fields (Q801599): Difference between revisions

Given a random field X on \({\mathbb{R}}^ d\) (i.e., a random element of the dual space of the set of infinitely differentiable functions with bounded support), X is in the domain of attraction of the random field \(\tilde X\) if for each \(\lambda >0\) there are constants \(C_{\lambda}>0\) and \(\alpha_{\lambda}\in {\mathbb{R}}\) such that the random fields \(X_{\lambda}(f)=\lambda^{-d}X(f(\cdot /\lambda))\) satisfy \(C_{\lambda}(X_{\lambda}-\alpha_{\lambda})\to^{d}\tilde X\), where \((X-\alpha)(f)=X(f)-\alpha \int f(x)dx\). Here \(\to^{d}\) signifies convergence of finite-dimensional distributions. As in related settings there exists a scaling exponent \(\theta\) such that \(C_{\lambda}=\lambda^{\theta}g(\lambda)\), where g is slowly varying at infinity, and \(\tilde X\) is self-similar with exponent \(\theta\). The authors' main result is that if \(Z_{\Lambda}\) is a Cox (or doubly stochastic Poisson or subordinated Poisson) random field directed by the stationary, ergodic, positive random field \(\Lambda\), if \(\Lambda\) admits scaling limit \({\tilde \Lambda}\) for the scaling parameters \(C_{\lambda}\), \(\alpha_{\lambda}\) and exponent \(\rho\), and if a technical condition is satisfied, then \(Z=Z_{\Lambda}\) has a scaling limit \(\tilde Z\) with parameters \(D_{\lambda},\beta_{\lambda}\) and exponent \(\theta\) that are calculated explicitly in terms of \(C_{\lambda},\alpha_{\lambda},\rho\). The form of the relationship among these parameters varies according as \(\rho <d/2\), \(\rho =d/2\) or \(\rho >d/2\). The technical condition alluded to above is that \(\Lambda ([0,1]^ d)\) admits an analytic moment generating function in some neighborhood of 0. [Since the authors deal with random fields it is unclear to the reviewer in what sense \(\Lambda ([0,1]^ d)\) exists.]